Articles
Case studies and reference implementations
Production work written up in full: what broke, how it was narrowed down, what the cause turned out to be, and what was put in place so it would not happen unnoticed again.
These intentionally omit proprietary source code, internal repository information, cloud project identifiers and confidential client data.
2026-10-11 · System Design · Elasticsearch · Kubernetes · Message Queue · Search
How Discord indexes trillions of messages
Discord replaced its Redis queue, split Elasticsearch into small cells and routed bulk writes by destination to operate search at trillion-message scale.
2026-10-11 · Payments · Ledger · System Design · Kafka · Data Consistency
Why an async payment platform still needs a strongly consistent ledger
Uber Gulfstream combines an asynchronous payment pipeline with immutable money orders, zero-sum validation and strongly consistent ledger balances.
2026-10-10 · System Design · Scalability · Caching · Database · Shopify
How Shopify handled 1.27 million requests per second during Black Friday
System-design lessons from BFCM 2022: distribute load, protect the database, cache repeated reads and prepare for traffic spikes.
2026-10-09 · MySQL · Database Partitioning · Reliability · System Design
Partition Relational Databases Before You Move Data
Lessons from GitHub's MySQL partitioning: schema domains, boundary linters, consistency checks, cutover mechanics, and production rollout.
2026-10-08 · MySQL · Inventory · Concurrency · System Design
What Shopify's MySQL Inventory Reservation Migration Teaches
Lessons from Shopify's Redis-to-MySQL migration: bounded row pools, lock design, connection visibility, and a safer rollout strategy.
2026-09-25 · Knowledge Base · Context Graph · MCP · tree-sitter · AI Agent · Docs-as-code
Docs-as-code Knowledge Base + Context Graph for a 15-project portfolio
Docs for 15 projects had drifted from the code. A docs-as-code KB, a context graph and an MCP server for AI agents, measured in 126 blind-graded sessions.
2026-09-18 · Next.js · Firebase Hosting · Cloud Run · SEO · Incident
How I Traced a Googlebot-only HTTP 500 Across Next.js, Cloud Run and the Hosting Edge
Users and every other crawler got 200 while Googlebot got 500 site-wide. The cause: a 404-path regression amplified by crawler verification at the hosting edge.
